http://akjamie.github.io/tags/durable-skills/ Recent content in Durable Skills on Jamie's Blog Hugo en-us Fri, 29 May 2026 00:00:00 +0000 http://akjamie.github.io/post/2026-05-29-self-improvement-agent-deep-dive/ Fri, 29 May 2026 00:00:00 +0000 http://akjamie.github.io/post/2026-05-29-self-improvement-agent-deep-dive/ <h1 id="the-self-improvement-agent-in-hermes-a-deep-dive">The Self-Improvement Agent in Hermes: A Deep Dive</h1> <blockquote> <p>Staff-engineer-level notes on Hermes Agent&rsquo;s procedural self-improvement loop: how the runtime turns solved problems, user corrections, and hard-won debugging paths into durable skills without interrupting the foreground task.</p> </blockquote> <hr> <blockquote> <p>[!NOTE]</p> <h3 id="executive-tldr">Executive TL;DR</h3> <p>Hermes self-improvement is not a magic <code>auto_create_skill()</code> function. It is a runtime pattern composed from five pieces:</p> <ul> <li><strong>Foreground guidance:</strong> the main agent is told to save or patch skills when a complex workflow succeeds.</li> <li><strong>Iteration counters:</strong> the runtime tracks how much tool-heavy work has happened since the last skill review.</li> <li><strong>Background review fork:</strong> after the user-visible answer is delivered, Hermes starts a quiet review agent with the conversation transcript.</li> <li><strong>Narrow tool whitelist:</strong> the review fork may call only memory and skill tools, not arbitrary shell, web, or browser tools.</li> <li><strong>Procedural memory tool:</strong> <code>skill_manage</code> writes, patches, deletes, and annotates skills under Hermes&rsquo; skill library.</li> </ul> <p>The architecture matters more than the prompt. The learning loop is isolated from the foreground answer, bounded by tool permissions, marked with provenance, and implemented through the same skill API the main agent can use.</p>